GitHub - DanielDeshmukh/autobots: Autobots | A decentralized model agentic swarm powered by NVIDIA NIM. Orchestrating high-precision, end-to-end authenticated software development through a specialized 6-file control architecture. Features automated security auditing, non-redundant task routing, and autonomous state management.

Autobots is a hierarchical multi-cluster coding swarm CLI that orchestrates multiple AI models to plan, implement, validate, and repair code against target repositories. It features a full tool system (Read, Write, Edit, Glob, Grep), interactive REPL mode, granular permissions, context management, hooks, MCP integration, and code review — all built on NVIDIA NIM models.

Features
Architecture
Installation
Quick Start
Projects Built by Autobots
Configuration
CLI Commands
Tool System
REPL Mode
Permissions
Context Management
Hooks & MCP
Code Review & Diagnostics
NVIDIA Models & Clusters
NVIDIA Skills
Execution Modes
Context Architecture
Cross-Platform Usage
Environment Variables
Troubleshooting
Development
License

Features

Core Swarm

Hierarchical Model Routing — 9 specialized clusters (Optimus, UltraMagnus, Jazz, RedAlert, Ratchet, Perceptor, Bumblebee, Ironhide, Wheeljack)
Phase-Based Execution — Inspect → Implement → Validate → Repair pipeline
Context Injection — Reads context/ files and injects project documentation into model prompts
Model Routing — Intelligent cluster selection based on task keywords and file extensions
Automatic Repair — Self-healing execution with validation-driven repair loops
Rollback Support — Snapshots before writes, autobots undo reverts changes
Response Streaming — Live character counter during model calls
Rich Status Output — Progress bars, estimated time, branch info via Rich

File Operations & Search

ReadTool — Read files with offset/limit, line numbers, binary/image support
WriteTool — Atomic writes (tmp + rename), directory creation, path sandboxing
EditTool — Targeted string replacement with exact match, replaceAll mode
GlobTool — Pattern-based file search (**/*.py, src/**/*.ts)
GrepTool — Regex content search across files, include filters
ToolRegistry — Register, dispatch, and manage tools

REPL & Interactive Mode

Interactive REPL — Conversational session with history and streaming
One-Shot Mode — autobots ask "question" for quick queries
Piped Input — cat file.py | autobots ask "explain this"
Slash Commands — /help, /clear, /cost, /compact, /model, /exit
Session Persistence — Save and restore conversations from JSON

Permissions & Safety

Tool-Level Permissions — Allow/deny/ask per tool call
Interactive Approval — y/n/a/Esc prompt for tool execution
Session Whitelisting — "Always allow" for trusted tools
Config Merge — Global (~/.autobots/settings.json) + project + env vars
Audit Logging — JSON-lines log of all permission decisions
Command Policy — Whitelist-based command validation, blocks dangerous patterns

Context Management

Token Estimation — len(text) // 4 with tiktoken fallback
Auto-Compaction — Summarizes conversation when approaching context limit
CLAUDE.md Hierarchy — Global → project → subdirectory instruction loading
Budget Tracking — Remaining context display, usage ratios

Hooks & Extensions

Pre/Post Tool Hooks — Run scripts or callbacks before/after tool execution
Abort on Failure — Hook failure stops the tool chain
MCP Integration — Connect to Model Context Protocol servers for custom tools
Command Hooks — Shell commands with env injection and 30s timeout

Code Review & Diagnostics

Git Diff Review — Analyze changes for large PRs, many files, patterns
Doctor Command — Health checks: API key, connectivity, Python version, package
PR Review — Review pull requests by number or diff

Workflow

Session Management — Durable checkpoints, resumable runs, audit trails
Configurable Modes — Supervised, milestone, or fully autonomous execution
Multi-Root File Writing — Write to src/, app/, lib/, tests/, docs/, scripts/
Validation Commands — Run tests, linters, and build commands automatically
Shell Completions — bash, zsh, and fish tab completion
Plugin System — before/after hooks for custom extensions
Skill Marketplace — Built-in packs for FastAPI, Django, React, Next.js
Web Dashboard — Real-time status on port 8080

Architecture

autobots/
├── cli.py                  # CLI entry point (22+ commands)
├── bootstrap.py            # Project profiling and context contract
├── config.py               # TOML + env config with validation
├── catalog.py              # Cluster definitions and model registry
├── workspace.py            # Target workspace safety and locking
├── errors.py               # Structured error classes
├── costs.py                # Token/cost estimation
├── git_utils.py            # Git auto-commit integration
├── preflight.py            # Doctor health checks
├── onboarding.py           # Interactive setup wizard
├── completions.py          # Shell tab completion
├── context_budget.py       # Context window management
├── plugins.py              # Plugin/hook system
├── marketplace.py          # Skill pack marketplace
├── dashboard.py            # Web dashboard
├── diff.py                 # Snapshot comparison
├── tools/                  # File operation tools
│   ├── base.py             # Tool ABC and ToolResult
│   ├── read.py             # ReadTool (offset/limit, binary)
│   ├── write.py            # WriteTool (atomic, sandboxed)
│   ├── edit.py             # EditTool (exact match, replaceAll)
│   ├── glob.py             # GlobTool (pattern search)
│   ├── grep.py             # GrepTool (regex search)
│   ├── registry.py         # ToolRegistry (dispatch)
│   ├── permissions.py      # PermissionRule, PermissionChecker
│   └── formatting.py       # Tool result formatting
├── repl/                   # Interactive REPL
│   ├── session.py          # ReplSession, Message, SessionStats
│   ├── commands.py         # SlashCommand registry
│   └── runner.py           # run_repl, run_one_shot, run_piped
├── permissions/            # Permission system
│   ├── settings.py         # PermissionSettings (global/project/env)
│   ├── logging.py          # PermissionLogger (audit trail)
│   └── interactive.py      # Interactive approval prompt
├── context/                # Context management
│   ├── manager.py          # ContextManager (token tracking)
│   ├── tokenizer.py        # estimate_tokens, count_tokens
│   └── claude_md.py        # CLAUDE.md hierarchy loader
├── hooks/                  # Hook system
│   └── manager.py          # HookManager, HookPoint, HookResult
├── mcp/                    # MCP integration
│   └── client.py           # MCPClient, MCPTool
├── review/                 # Code review
│   ├── git_review.py       # Diff parsing, review
│   └── diagnostics.py      # Doctor checks
├── router/                 # Swarm orchestration
│   ├── core.py             # Main routing orchestration
│   ├── models.py           # Data models
│   ├── planning.py         # Cluster assignment
│   ├── stages.py           # Stage execution
│   └── phases.py           # Phase reading
├── executor/               # Execution engine
│   ├── autonomy.py         # Autonomous loop
│   ├── modes.py            # Execution modes
│   ├── state.py            # Session state and rollback
│   ├── commands.py         # Command validation
│   ├── gate.py             # Test-then-commit gate
│   └── validation.py       # Validation handling
├── planning/               # Plan generation
│   ├── core.py             # Plan entry point
│   ├── scanner.py          # Repository scanning
│   └── synthesis.py        # Phase synthesis
├── skills/                 # Skill injection
│   ├── loader.py           # Skill pack loading
│   ├── cluster_prompts.py  # Cluster-specific prompts
│   └── nvidia/             # 17 NVIDIA skill files
└── utils/
    └── retry.py            # Exponential backoff

Installation

Prerequisites

Python 3.11 or higher
NVIDIA API Key (for model execution)

PyPI Install

pip install autobot-swarm

Development Install

git clone https://github.com/DanielDeshmukh/autobots.git
cd autobots
python -m pip install -e . --no-build-isolation

Quick Start

# 1. Set your API key
$env:NVIDIA_API_KEY = "your_key_here"

# 2. Navigate to your project
cd C:\path\to\your\project

# 3. Initialize (creates context/ directory)
autobots init

# 4. Generate a plan
autobots plan --goal "Add user authentication"

# 5. Run the swarm
autobots run --supervised

Projects Built by Autobots

Real projects generated entirely by the multi-model swarm — no manual code, no hand-holding. Each project starts from a single sentence and goes through planner → workers → repair pipeline.

Todo App

"Build a todo app with React, TypeScript, and Vite. Features: add todos, toggle complete, delete, filter (all/active/completed), counter showing remaining."

Detail	Value
Stack	React 19, TypeScript, Vite
Files Generated	23
Models Used	2 (`qwen3-next-80b`, `llama-3.3-70b`)
Workers	10 parallel (semaphore-limited to 2 concurrent)
Design	Dark theme, glass morphism, gradient accents
Features	Add, delete, toggle complete, filter (all/active/completed), counter

How it was built:

Planner (qwen3-next-80b) decomposed the goal into 10 subtasks — types, hooks, components (Input, Item, Filter, Counter, List), styles, tests
Critical file injection added index.html, package.json, main.tsx, index.css automatically (planner forgets these ~80% of the time)
Workers executed in parallel with shared context — each model built 1-2 files with type contracts ensuring all components use the same interfaces
Repair phase detected missing files from timed-out workers and regenerated them, then ran TypeScript compilation check to fix any remaining type mismatches
Import path auto-fix resolved wrong import paths between components

Repo: github.com/DanielDeshmukh/todo-by-autobots

Memory Card Flip

"Build a memory card flip game. Match pairs to win. Features: card flip animation, theme selection (animals/flags/emojis), difficulty levels, timer, move counter, and win detection."

Detail	Value
Stack	React 19, TypeScript, Vite, Tailwind CSS, Zustand
Design System	NVIDIA DESIGN.md (black canvas, green accent)
Pipeline	Mark II — Dual-Model Collaboration
Models Used	2 (`qwen3.5-122b` generator, `qwen3-next-80b` reviewer)
Workers	Sequential (1 at a time due to rate limits)
Features	Card flip animation, 3 themes, 3 difficulties, timer, moves, matches, progress bar

How it was built:

Planner (qwen3-next-80b) decomposed the goal into 7 subtasks — store, components (Card, GameGrid, StatsBar, ProgressBar, ThemeSelector, DifficultySelector, Celebration), styles, types, constants
Critical file injection added index.html, package.json, main.tsx, index.css, tsconfig.json automatically
Dual-model workers — each subtask gets a generator (qwen3.5-122b) and reviewer (qwen3-next-80b). Generator creates code, reviewer checks for bugs before writing
Repair phase detected missing files and ran TypeScript validation to fix type mismatches
Design system injection — NVIDIA DESIGN.md was injected into prompts so all components use black (#000000) canvas with green (#76B900) accent, angular 2px radius

Problems faced during generation:

Problem	Root Cause	Fix
Timer showed `NaN:NaN`	gameStore used `timer`, StatsBar read `timeElapsed`	Aligned property names across components
Cards rendered as tiny squares	framer-motion not installed, no explicit sizing	Replaced with pure CSS flip animations
Cards didn't flip on click	`initializeGame()` never called on mount	Added `useEffect` to call it on mount
6x6 showed 32 cards instead of 36	Theme had 16 items, 6x6 needs 18 unique	Increased themes to 18+ items with cycling
Missing PostCSS config	`postcss.config.js` incompatible with ESM	Renamed to `.cjs` extension
Cross-file contract mismatches	Parallel workers generated independently	Switched to single-worker with full context

Repo: github.com/DanielDeshmukh/memory-card-flip

Configuration

API Key Setup

# Option 1: .env file
echo NVIDIA_API_KEY=your_key_here > .env

# Option 2: Environment variable
$env:NVIDIA_API_KEY = "your_key_here"

Configuration File

Create .autobots.toml or autobots.toml in your project root or $HOME:

[autobots]
model_selection_profile = "balanced"   # balanced, speed, or quality
parallel_planning = false
disable_live_catalog = false
safety_branch = "autobots-safety"
default_mode = "supervised"            # supervised, milestone, autonomous
milestone_threshold = 3
max_verification_attempts = 3
temperature = 0.2
max_tokens = 4096

# Optional: custom model registry
# model_registry_path = "./my-registry.json"

# Optional: extra custom clusters
# [autobots.extra_clusters]
# MyCluster = ["nvidia/custom-model-1"]

Permission Settings

Create ~/.autobots/settings.json (global) or .autobots/settings.json (project):

{
  "permissions": {
    "default": "ask",
    "rules": [
      {"tool_pattern": "Read", "permission": "allow"},
      {"tool_pattern": "Glob", "permission": "allow"},
      {"tool_pattern": "Grep", "permission": "allow"},
      {"tool_pattern": "Write", "permission": "ask"},
      {"tool_pattern": "Edit", "permission": "ask"},
      {"tool_pattern": "Bash", "permission": "ask"}
    ]
  }
}

CLI Commands

Command	Description
`autobots init`	Check context files in target project
`autobots init --interactive`	Interactive setup wizard
`autobots plan`	Generate implementation roadmap
`autobots plan --goal "text"`	Plan with specific goal
`autobots plan --append`	Append to existing plan
`autobots plan --dry-run`	Preview without writing
`autobots run [task]`	Execute phases with autonomy mode
`autobots run --supervised`	Manual approval per phase (default)
`autobots run --milestone`	Approval every N phases
`autobots run --autonomous`	Fully autonomous execution
`autobots resume`	Resume from last checkpoint
`autobots engage`	Interactive mode with startup screen
`autobots ask "question"`	One-shot question answering
`autobots steer "instruction"`	Add mid-task steering instructions
`autobots status`	Rich status with progress bars
`autobots explain <id>`	Show audit trail for a phase/task
`autobots stats`	Usage statistics and costs
`autobots undo`	Rollback to previous snapshot
`autobots snapshots`	List available snapshots
`autobots diff`	Compare workspace to snapshot
`autobots logs`	View audit trail
`autobots doctor`	Preflight health checks
`autobots catalog`	Browse NVIDIA model registry
`autobots config validate`	Validate TOML configuration
`autobots completions`	Generate shell completions
`autobots marketplace`	Browse skill packs
`autobots dashboard`	Launch web dashboard
`autobots validate-models`	Test NVIDIA API connectivity
`autobots publish`	Build and publish to PyPI

Tool System

Autobots includes a full file operation tool system:

ReadTool

# Read with line numbers
ReadTool.run(file_path="/path/to/file.py")

# Read with offset and limit
ReadTool.run(file_path="/path/to/file.py", offset=10, limit=50)

WriteTool

# Atomic write (creates parent directories automatically)
WriteTool.run(file_path="/path/to/file.py", content="print('hello')")

# Sandboxed write (only allowed paths)
WriteTool(allowed_paths=["/project/src"]).run(file_path="...", content="...")

EditTool

# Exact string replacement
EditTool.run(file_path="main.py", old_string="old_code", new_string="new_code")

# Replace all occurrences
EditTool.run(file_path="main.py", old_string="foo", new_string="bar", replace_all=True)

GlobTool

# Find files by pattern
GlobTool.run(pattern="**/*.py", path="/project")

# Find TypeScript files
GlobTool.run(pattern="src/**/*.tsx", path="/project")

GrepTool

# Regex search
GrepTool.run(pattern=r"def \w+", path="/project/src")

# Search specific file types
GrepTool.run(pattern="TODO", path="/project", include="*.py")

REPL Mode

Interactive Session

autobots
# Autobots REPL (model: meta/llama-3.1-8b-instruct)
# Type /help for commands, /exit to quit.
#
# You> What is this project about?
# Assistant: This project is...

One-Shot Mode

autobots ask "What does the main.py file do?"

Piped Input

cat main.py | autobots ask "Explain this code"

Slash Commands

Command	Description
`/help`	Show available commands
`/clear`	Clear conversation history
`/cost`	Show token usage and estimated cost
`/compact`	Summarize conversation to save tokens
`/model <name>`	Switch model mid-session
`/exit`	Exit the REPL

Permissions

Permission Levels

allow — Auto-approved, no prompt
ask — Interactive approval required
deny — Blocked, cannot execute

Interactive Approval

When a tool requires approval:

Allow Write?
  file_path: /project/src/main.py
  content: print('hello')
  [y]es / [n]o / [a]lways / [Esc] cancel

y — Approve this once
n — Deny this call
a — Always allow this tool for the session
Esc — Cancel

Configuration

Global (~/.autobots/settings.json):

{
  "permissions": {
    "default": "ask",
    "rules": [
      {"tool_pattern": "Read", "permission": "allow"},
      {"tool_pattern": "Write", "permission": "ask"}
    ]
  }
}

Project (.autobots/settings.json):

{
  "permissions": {
    "rules": [
      {"tool_pattern": "Bash(npm test)", "permission": "allow"},
      {"tool_pattern": "Bash(rm *)", "permission": "deny"}
    ]
  }
}

Environment Variables:

$env:AUTOBOTS_ALLOWED_TOOLS = "Read,Glob,Grep"
$env:AUTOBOTS_DENIED_TOOLS = "Bash(rm *)"

Context Management

Token Tracking

Tracks token usage per message
Estimates remaining context budget
Auto-compacts at 80% threshold

CLAUDE.md Hierarchy

Instructions loaded from multiple locations:

~/.autobots/CLAUDE.md           (global)
./CLAUDE.md                     (project root)
./src/CLAUDE.md                 (subdirectory)

Compaction

You> /compact
Compacted 24 messages into summary.

Hooks & MCP

Pre/Post Tool Hooks

Register callbacks that run before or after tool execution:

from autobots.hooks import HookManager, Hook, HookPoint

mgr = HookManager()
mgr.register(Hook(
    name="lint",
    point=HookPoint.POST_TOOL,
    command="ruff check {file_path}"
))

MCP Integration

Connect to Model Context Protocol servers:

from autobots.mcp import MCPClient

client = MCPClient(command="npx", args=["-y", "@modelcontextprotocol/server-filesystem"])
client.connect()
tools = client.list_tools()
result = client.call_tool("read_file", {"path": "/etc/hosts"})

Code Review & Diagnostics

Git Diff Review

# Review current changes
autobots review

# Review PR by number
autobots pr-review 42

Doctor Command

autobots doctor
# Autobots Doctor Results:
#   [PASS] Python Version: Python 3.13 (OK)
#   [PASS] API Key: NVIDIA_API_KEY set (ends with ...abc1)
#   [PASS] Connectivity: NVIDIA API reachable
#   [PASS] Package: autobot-swarm installed
#   4/4 checks passed

NVIDIA Models & Clusters

Cluster	Role	Models	Keywords
Optimus	Command & Routing	9	plan, roadmap, phase, orchestrate
UltraMagnus	Backend & Architecture	12	backend, api, database, service
RedAlert	Security & Safety	10	security, auth, guardrail, validation
Jazz	Frontend & Creative	11	ui, ux, css, image, visual
Ratchet	Debug & Repair	10	debug, fix, refactor, patch, repair
Perceptor	Retrieval & Parsing	11	ocr, rag, embedding, document
Bumblebee	Communication & Media	9	speech, voice, translation, audio
Ironhide	Physical & Simulation	8	simulation, physics, autonomous
Wheeljack	Scientific Specialist	7	science, molecule, protein, quantum

NVIDIA Skills

Tier 1 (Always Loaded)

Skill	Cluster(s)	Description
`agent-skills.md`	Optimus, UltraMagnus	NemoClaw agent architecture
`autonomous-agent-research.md`	Optimus, Wheeljack	Autonomous research workflow
`session-memory.md`	Optimus, Jazz	Durable session memory
`skill-evolution.md`	ALL	Self-improvement protocol
`safety-policy.md`	Ironhide	Safety taxonomy
`rag-blueprint.md`	UltraMagnus, Optimus	RAG architecture
`rag-eval.md`	Jazz	RAG evaluation metrics
`dynamo-deployment.md`	UltraMagnus, Wheeljack	Dynamo K8s recipes
`dynamo-router.md`	UltraMagnus, Optimus	Dynamo router modes
`retrieval.md`	UltraMagnus, Bumblebee	Document retrieval pipeline

Tier 2 (Conditional)

Skill	Loaded When	Description
`nemotron-customize.md`	fine-tuning, training	Nemotron customization
`cuopt-routing.md`	routing, scheduling	cuOpt vehicle routing
`cuopt-optimization.md`	optimization, LP/MILP/QP	cuOpt numerical optimization
`cudf.md`	pandas, dataframe	cuDF GPU DataFrames
`neautomodel-recipe.md`	training at scale	Distributed training
`kubernetes-infra.md`	Kubernetes, K8s	Physical AI infra
`holoscan.md`	video, edge AI	Holoscan video analytics
`cudaq.md`	quantum computing	CUDA-Q quantum simulation

Execution Modes

Supervised (Default)

autobots run --supervised

Requires approval before each phase. Full human control.

Milestone

autobots run --milestone

Requires approval after N phases (configurable via milestone_threshold).

Autonomous

autobots run --autonomous

No approval gates. Fastest execution.

Context Architecture

Autobots expects six context files under context/:

File	Purpose
`architecture.md`	Project architecture and structure
`roadmap.md`	Phase definitions with goals and validation
`progress-tracker.md`	Phase status tracking
`project-briefing.md`	Project overview and goals
`security-auth.md`	Security and authentication notes
`ui-components.md`	UI component documentation

Cross-Platform Usage

Windows

pip install autobot-swarm
$env:NVIDIA_API_KEY = "your_key"
cd C:\path\to\your\project
autobots init
autobots plan --goal "Add feature X"
autobots run --supervised

macOS / Linux

pip install autobot-swarm
export NVIDIA_API_KEY="your_key"
cd /path/to/your/project
autobots init
autobots plan --goal "Add feature X"
autobots run --supervised

Environment Variables

Variable	Description	Default
`NVIDIA_API_KEY`	NVIDIA API key for model access	Required
`AUTOBOTS_MODEL_SELECTION_PROFILE`	Model selection: balanced/speed/quality	balanced
`AUTOBOTS_ENABLE_PARALLEL_PLANNING`	Enable parallel workstreams	false
`AUTOBOTS_DISABLE_LIVE_CATALOG`	Use bundled models only	false
`AUTOBOTS_SAFETY_BRANCH`	Required git branch name	autobots-safety
`AUTOBOTS_DEFAULT_MODE`	Default execution mode	supervised
`AUTOBOTS_MILESTONE_THRESHOLD`	Phases per approval	3
`AUTOBOTS_MAX_VERIFICATION_ATTEMPTS`	Retry limit	3
`AUTOBOTS_ALLOWED_TOOLS`	Comma-separated allowed tools	(none)
`AUTOBOTS_DENIED_TOOLS`	Comma-separated denied tools	(none)

Troubleshooting

Error	Solution
`NVIDIA_API_KEY is missing`	Set `NVIDIA_API_KEY` in `.env` or environment variable
`Missing: roadmap.md`	Run `autobots plan` to generate context files
`Switch to autobots-safety branch`	`git checkout -b autobots-safety`
`Command not in safety whitelist`	Autobots blocks dangerous commands. Use safe alternatives
`No checkpoint found`	Start fresh with `autobots run`
`old_string not found`	Provide exact surrounding lines for unique match
`old_string found N times`	Add more context or use `replace_all=true`

Development

Running Tests

# All tests
python -m pytest tests/ -v

# Specific test file
python -m pytest tests/test_tools.py -v

# With coverage
python -m pytest tests/ --cov=autobots --cov-report=html

Test Suite

Test File	Tests	Coverage
`test_tools.py`	46	File operations (Read, Write, Edit, Glob, Grep, Registry, Permissions, Formatting)
`test_tools_integration.py`	5	Integration: read → edit → verify → read
`test_repl.py`	20	REPL session, commands, runner
`test_permissions.py`	12	Permission settings, logging, merge
`test_context.py`	18	Token estimation, compaction, CLAUDE.md
`test_hooks_mcp.py`	13	Hooks, MCP client
`test_review.py`	12	Git review, diagnostics
`test_final_decision.py`	10	Release readiness
Total	136

License

MIT License — See LICENSE for details.

Acknowledgments

Built with NVIDIA NIM for model access
Powered by Rich for terminal UI
Uses OpenAI compatible API interface

Name		Name	Last commit message	Last commit date
Latest commit History 242 Commits
.claude		.claude
.github/workflows		.github/workflows
assets		assets
autobots		autobots
context		context
scripts		scripts
tests		tests
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
COST_VERIFICATION.md		COST_VERIFICATION.md
CROSS_PLATFORM.md		CROSS_PLATFORM.md
E2E_JOURNEYS.md		E2E_JOURNEYS.md
KNOWN_ISSUES.md		KNOWN_ISSUES.md
LICENSE		LICENSE
MARK_II_PLAN.md		MARK_II_PLAN.md
NVIDIA_MODELS.md		NVIDIA_MODELS.md
PARITY_ROADMAP.md		PARITY_ROADMAP.md
PROJECTS.md		PROJECTS.md
README.md		README.md
STATUS.md		STATUS.md
TRUST_TESTING.md		TRUST_TESTING.md
autobots.toml.example		autobots.toml.example
nvidia_models.json		nvidia_models.json
package-description.md		package-description.md
pyproject.toml		pyproject.toml
setup.cfg		setup.cfg

Folders and files

Latest commit

History

Repository files navigation

Table of Contents

Features

Core Swarm

File Operations & Search

REPL & Interactive Mode

Permissions & Safety

Context Management

Hooks & Extensions

Code Review & Diagnostics

Workflow

Architecture

Installation

Prerequisites

PyPI Install

Development Install

Quick Start

Projects Built by Autobots

Todo App

Memory Card Flip

Configuration

API Key Setup

Configuration File

Permission Settings

CLI Commands

Tool System

ReadTool

WriteTool

EditTool

GlobTool

GrepTool

REPL Mode

Interactive Session

One-Shot Mode

Piped Input

Slash Commands

Permissions

Permission Levels

Interactive Approval

Configuration

Context Management

Token Tracking

CLAUDE.md Hierarchy

Compaction

Hooks & MCP

Pre/Post Tool Hooks

MCP Integration

Code Review & Diagnostics

Git Diff Review

Doctor Command

NVIDIA Models & Clusters

NVIDIA Skills

Tier 1 (Always Loaded)

Tier 2 (Conditional)

Execution Modes

Supervised (Default)

Milestone

Autonomous

Context Architecture

Cross-Platform Usage

Windows

macOS / Linux

Environment Variables

Troubleshooting

Development

Running Tests

Test Suite

License

Acknowledgments

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks